Posted by Jogi from joe154.pp.dlc.fi (195.218.77.154) on March 07, 1998 at 07:40:45:
I just compiled RRBench under VC 5.0, (no modification needed) and it
seems that VC is producing quite faster code than LCC.
Under LCC i did not get LCC:s optimations to work, RRB just crashed
in first 3d objext test.
I changed V_IssueCmdBufferASync() to V_ResetCmdBuffer() to
see how fast the RRB code is. (it is not supposed to be fast,
as it is benchmark :-)
VC code was 50-100% faster than LCC code, and even then code was
horrible slow, atleast for game engine..
It seems that float calculations are somehow slow and converting
float to rredline fixedpoint integers is also slow thing to do.
So if anyone knows fast float to fixedint conversion routine, please share
you knowledge to us.
(this is float to int define that comes with rredline)
#define FLTOIFIX(f) ( (v_u32)( ((float)(f)) * 65536.0) )
One solution could be that that float value is from start kept
f*65536 so there would be only cast to integer (->v_u32)
But will it affect accuracy?